Pre-alpha. Prototyping at home.

A voice assistant that waits for every word.

Project Noah is a home voice assistant for people whose speech doesn’t fit what mainstream assistants expect. It started with my son, who has childhood apraxia of speech.

Get in touch See how it listens
Animation: a voice waveform says “wab”, then pauses for a long time, then says “bit”. Many assistants would stop listening during the pause. Project Noah keeps listening, joins the two parts and understands the word “rabbit”.

Voice tech expects quick, tidy speech. Plenty of people don’t talk that way.

Noah is 11. He has childhood apraxia of speech (CAS), a motor planning disorder: he knows the word he wants, but his mouth can’t always produce the sequence of sounds cleanly, and the same word can come out differently from one try to the next. CAS affects about 1 in 1,000 children ages 4 to 8, according to ASHA.

The voice assistants in our house time out, mishear and interrupt him. Each failed request is a small reminder that the device wasn’t built with him in mind. After a while, he stopped asking it for things.

Project Noah started as a single OpenHome DevKit on our kitchen table. The goal is a home assistant that is patient by design, for CAS first and then for stuttering, dysarthria, AAC users and anyone else whose speech doesn’t fit the template.

Brennan Decker, Noah’s dad

Built around how people actually talk.

  • Long pauses are okay.

    Silence isn’t failure. Project Noah is designed to wait through the pauses that make other assistants give up, so the speaker gets to finish the thought.

  • “Wabbit” still means rabbit.

    Speech recognition tuned for substitutions, dropped syllables and words that sound different each time, with context filling in what articulation can’t.

  • It never corrects.

    Pronunciation work belongs to the speech therapist. The assistant listens, understands and answers, and it doesn’t turn a request into a lesson.

  • A wake word you can say.

    “Hey Google” starts with a hard /g/. Each household picks a wake word built from sounds the speaker produces reliably.

  • Voices stay home.

    The goal is speech processing on the device, with any data sharing opt-in and controlled by the family. We’re honest below about where that stands today.

  • Clinically informed.

    Conversation design shaped with speech-language pathologists who work with motor speech disorders. A clinical advisory group is being formed now.

How it runs at home

  1. A small device listens

    Project Noah runs on the OpenHome DevKit, a Raspberry Pi based voice device with a developer SDK, placed wherever the family wants it.

  2. It waits, then understands

    Patient end-of-speech timing and approximation-tolerant matching turn “wab… bit” into “rabbit” instead of an error.

  3. The family stays in control

    A parent or caregiver owns the account, picks the wake word and decides what, if anything, is ever shared.

Where privacy stands today. The standard OpenHome setup sends speech to cloud transcription and voice providers (Deepgram or ElevenLabs, per OpenHome’s docs). A fully on-device speech path is a roadmap goal, not a shipped feature, and this page will say so until it is.

Who it’s for

People whose speech is understood by the people who love them, and by almost no device.

People with speech differences
Starting with childhood apraxia of speech, then stuttering, dysarthria and AAC users.
Parents and caregivers
The adults who set up the device, own the account and decide how it’s used at home.
Speech-language pathologists
Clinicians who want a home tool that complements therapy and never replaces it.

About children. Project Noah started for a child, and children are who we most want to serve. That’s exactly why we’re moving carefully: before any child uses it, the hardware and platform terms, verifiable parental consent and a privacy review all have to be right.

Roadmap

Status: pre-alpha. There is no public release yet.

  1. Now

    Prototype at home

    Early prototyping on one OpenHome DevKit, this site, and forming a clinical advisory group.

  2. Next

    On-device research

    Document the patient-listening settings and research a fully on-device speech path.

  3. Then

    Family alpha

    A small alpha with families, opened only once consent, privacy review and hardware terms are in place.

Get involved

No forms, no sign-ups. Pick the door that fits and it opens your own email app.

Questions

Is this a replacement for speech therapy?

No, and it never will be. Project Noah is meant to be a patient conversation partner at home that complements the work someone does with their speech-language pathologist. It makes no therapeutic claims.

Can I use it today?

Not yet. Project Noah is pre-alpha. If you email us, we’ll let you know when there is something you can try.

What data does it collect?

This website collects nothing: no forms, cookies, analytics or trackers. For the assistant, the design goal is on-device speech processing, with any data used to improve models opt-in and controlled by the family. Today the standard OpenHome pipeline uses cloud speech providers, so we don’t claim on-device processing yet.

Is Project Noah made by OpenHome?

No. Project Noah is an independent project built on OpenHome’s DevKit and developer platform.

Will it work for speech differences other than CAS?

The first version is tuned for CAS, but the approach (patient timing, tolerance for approximations, custom wake words, no correction) is meant to help anyone whose speech doesn’t fit the template, including stuttering, dysarthria and AAC users.

Who is building this?

Brennan Decker, Noah’s dad and a product leader who has spent the last decade building trust, identity and personalization systems at companies like Ticketmaster, Bluecore, Fanatics and Finish Line. Brennan on LinkedIn.